This article uses default data from 2014 to 2021 to study the recovery rate of credit bonds in China. The average recovery rate one year after bond default (as of the end of 2021) is 8% (11%), with an asymmetric bipolar distribution, concentrated on both sides of 0% and 100%. With different horizon, the gold recovery horizon for defaulting bonds is one year. The proportion of Existing bonds in assets is an important factor affecting the recovery rate. Besides, with or without guarantor, issued in the inter-bank market or not, state owned enterprise or not, publicly listed or not, short-term treasury rate and spread over long and short treasury rate significantly affect the default recovery rate. In terms of recovery rate prediction, machine learning models are significantly superior to linear regression models. Among them, the prediction of ensemble learning model is the best.